DiscoStyle: Multi-level Logistic Ranking for Personalized Image Style Preference Inference

Zhen-Wei He, Lei Zhang, Fang-Yi Liu. DiscoStyle: Multi-level Logistic Ranking for Personalized Image Style Preference Inference. International Journal of Automation and Computing, 2020, 17(5): 637-651. doi: 10.1007/s11633-020-1244-1
 Citation: Zhen-Wei He, Lei Zhang, Fang-Yi Liu. DiscoStyle: Multi-level Logistic Ranking for Personalized Image Style Preference Inference. International Journal of Automation and Computing, 2020, 17(5): 637-651.

## DiscoStyle: Multi-level Logistic Ranking for Personalized Image Style Preference Inference

###### Author Bio: Zhen-Wei He received B. Eng. degree in information engineering from Tianjin University, China in 2014. From July 2014 to June 2016, he worked in Chongqing Cable Network Inc., China. Now, he is a Ph. D degree candidate in Chongqing University, China. His research interests include deep learning and computer vision.E-mail: hzw@cqu.edu.cnORCID iD: 0000-0002-6122-9277 Lei Zhang received the Ph. D degree in circuits and systems from the College of Communication Engineering, Chongqing University, China in 2013. He worked as a Post-Doctoral Fellow with Hong Kong Polytechnic University, China from 2013 to 2015. He is currently a professor/distinguished research fellow with Chongqing University, China. He has authored more than 90 scientific papers in top journals and top conferences. He serves as associate editors for IEEE Transactions on Instrumentation and Measurement, Neural Networks, etc. He is a senior member of IEEE. His research interests include machine learning, pattern recognition, computer vision and intelligent systems. E-mail: leizhang@cqu.edu.cn (Corresponding author)ORCID iD: 0000-0002-5305-8543 Fang-Yi Liu received the B. Eng. degree in communication engineering from Guangxi University, China in 2017. Since September 2017, he is a master student in information and communication engineering in Chongqing University, China. His research interests include person re-identification and deep learning.E-mail: fangyiliu@cqu.edu.cnORCID iD: 0000-0001-8815-0254
• Figure  1.  User-specific facial preference reasoning and prediction tasks, which shows the user′ s preference reasoning and recommendation to female faces. Notably, the anchors represent the preferred faces selected by users. Color versions of the figures in this paper are available online.

Figure  2.  The paradigm for our PFR method for feature representation, which shows the ROI detection and alignment, appearance feature extraction with three local regions in red (eye part), green (nose part) and blue (lip-chin part), and the geometric landmark coordinate feature vector (shape). In total, five kinds of features including 1 geometric feature, 3 local appearance features and 1 global appearance feature.

Figure  3.  The schematic of our proposed ONSS approach, which shows how to select the potential negative samples (i.e., non-preferred faces) on-line without traversing all the gallery faces, because in real-application the gallery faces are infinity. In this figure, the gallery face (red square) is recognized to be potential negative sample, because it shows the maximum distance to the center of positive anchors in red circle with a higher proportion (3/5 > 0.5). That is, the gallery face is recognized to be negative sample under 3 feature modalities rather than 2. Notably, for balancing between positive samples (i.e., preferred faces or anchors) and negative samples (i.e., non-preferred faces), the ONSS program is automatically stopped when the number of selected negative samples achieves to the number of anchors.

Figure  4.  The flowchart of our MLR model. The geometric features and appearance features are fed into the logistic regression model respectively in the $1\rm{st}$ LR level. The global appearance features are also used for SVM and MLP to get 3-style score and 5-style score vector. All results of the $1\rm{st}$ LR level are then fed into the $2\rm{nd}$ LR level, and we can get the final preference score of the input face.

Figure  5.  Examples of the preferred faces from two users in our StyleFace dataset. For each user, 10 anchors (the first row) are used for training and the remaining 30 faces (the last three rows) are used for testing. All images have been detected and aligned.

Figure  6.  The PR curves of all the compared methods based on three evaluation protocols. The $\rm{1st}$ and the $\rm{2nd}$ rows show the PR curves of female and male, respectively. The online color figure can get a better view.

Figure  7.  The feature distribution of training set, high-score samples and low-score samples for female (a) and male (b). The numbers below the red and green bounding boxes denote the rankings of the preference scores predicted by the proposed DiscoStyle. The online color figure can get a better view.

Figure  8.  Visualization of the anchors, recommended faces and non-recommended faces for female and male data in LFW dataset. For each user, the $\rm{1st}$ row shows the training samples (i.e., preferred anchors) selected by the user, the $\rm{2nd}$ row shows the recommended faces (most probably preferred faces) with the highest preference scores, and the $\rm{3rd}$ row shows the faces (most probably non-preferred faces) with the lowest preference scores by using the proposed DiscoStyle method.

•  [1] Azzedine Yahiaoui.  A Practical Approach to Representation of Real-time Building Control Applications in Simulation . International Journal of Automation and Computing, 2020, 17(3): 464-478. doi: 10.1007/s11633-018-1131-1 [2] Dilantha Haputhanthri, Gunavaran Brihadiswaran, Sahan Gunathilaka, Dulani Meedeniya, Sampath Jayarathna, Mark Jaime, Christopher Harshaw.  Integration of Facial Thermography in EEG-based Classification of ASD . International Journal of Automation and Computing, 2020, 17(): 1-18. doi: 10.1007/s11633-020-1231-6 [3] Kittinun Aukkapinyo, Suchakree Sawangwong, Parintorn Pooyoi, Worapan Kusakunniran.  Localization and Classification of Rice-grain Images Using Region Proposals-based Convolutional Neural Network . International Journal of Automation and Computing, 2020, 17(2): 233-246. doi: 10.1007/s11633-019-1207-6 [4] Shui-Guang Tong, Yuan-Yuan Huang, Zhe-Ming Tong.  A Robust Face Recognition Method Combining LBP with Multi-mirror Symmetry for Images with Various Face Interferences . International Journal of Automation and Computing, 2019, 16(5): 671-682. doi: 10.1007/s11633-018-1153-8 [5] Ao-Xue Li, Ke-Xin Zhang, Li-Wei Wang.  Zero-shot Fine-grained Classification by Deep Feature Learning with Semantics . International Journal of Automation and Computing, 2019, 16(5): 563-574. doi: 10.1007/s11633-019-1177-8 [6] Snehasis Banerjee, Tanushyam Chattopadhyay, Utpal Garain.  A Wide Learning Approach for Interpretable Feature Recommendation for 1-D Sensor Data in IoT Analytics . International Journal of Automation and Computing, 2019, 16(6): 800-811. doi: 10.1007/s11633-019-1185-8 [7] Xiang Zhang, Qiang Yang.  Transfer Hierarchical Attention Network for Generative Dialog System . International Journal of Automation and Computing, 2019, 16(6): 720-736. doi: 10.1007/s11633-019-1200-0 [8] Fadhlan Kamaru Zaman, Amir Akramin Shafie, Yasir Mohd Mustafah.  Robust Face Recognition Against Expressions and Partial Occlusions . International Journal of Automation and Computing, 2016, 13(4): 319-337. doi: 10.1007/s11633-016-0974-6 [9] Liu-Yuan Chen, Jie Yang, Guo-Guo Xu, Yun-Qing Liu, Jun-Tao Li, Cun-Shuan Xu.  Biomarker Identification of Rat Liver Regeneration via Adaptive Logistic Regression . International Journal of Automation and Computing, 2016, 13(2): 191-198. doi: 10.1007/s11633-015-0919-5 [10] Danasingh Asir Antony Gnana Singh, Subramanian Appavu Alias Balamurugan, Epiphany Jebamalar Leavline.  An Unsupervised Feature Selection Algorithm with Feature Ranking for Maximizing Performance of the Classifiers . International Journal of Automation and Computing, 2015, 12(5): 511-517. doi: 10.1007/s11633-014-0859-5 [11] De-Rong Liu, Hong-Liang, Li Ding Wang.  Feature Selection and Feature Learning for High-dimensional Batch Reinforcement Learning: A Survey . International Journal of Automation and Computing, 2015, 12(3): 229-242. doi: 10.1007/s11633-015-0893-y [12] Hai-Shun Du, Qing-Pu Hu, Dian-Feng Qiao, Ioannis Pitas.  Robust Face Recognition via Low-rank Sparse Representation-based Classification . International Journal of Automation and Computing, 2015, 12(6): 579-587. doi: 10.1007/s11633-015-0901-2 [13] Majda Ltaief, Anis Messaoud, Ridha Ben Abdennour.  Optimal Systematic Determination of Models' Base for Multimodel Representation: Real Time Application . International Journal of Automation and Computing, 2014, 11(6): 644-652. doi: 10.1007/s11633-014-0815-4 [14] Paramita Guha, Sunita Mishra.  Reduced Order Modeling & Controller Design for Mass Transfer in a Grain Storage System . International Journal of Automation and Computing, 2014, 11(4): 399-403. doi: 10.1007/s11633-014-0805-6 [15] Li Wang, Rui-Feng Li, Ke Wang, Jian Chen.  Feature Representation for Facial Expression Recognition Based on FACS and LBP . International Journal of Automation and Computing, 2014, 11(5): 459-468. doi: 10.1007/s11633-014-0835-0 [16] .  Methods for Controlling the Authenticity of Textual Information Transfer on the Basis of Statistical and Structural Redundancy . International Journal of Automation and Computing, 2012, 9(5): 518-529. doi: 10.1007/s11633-012-0675-8 [17] Wei-Guo Yi, Ming-Yu Lu, Zhi Liu.  Regression Analysis of the Number of Association Rules . International Journal of Automation and Computing, 2011, 8(1): 78-82. doi: 10.1007/s11633-010-0557-x [18] .  Efficient Graph-based Genetic Programming Representation with Multiple Outputs . International Journal of Automation and Computing, 2008, 5(1): 81-89. doi: 10.1007/s11633-008-0081-4 [19] Clemens Dietl, Uwe K. Rakowsky.  An Operating Strategy for High-availability Multi-station Transfer Lines . International Journal of Automation and Computing, 2006, 3(2): 125-130. doi: 10.1007/s11633-006-0125-6 [20] Cheng-Wei Dai,  Shuang-Hua Yang,  Roger Knott.  Data Transfer Over the Internet for Real Time Applications . International Journal of Automation and Computing, 2006, 3(4): 414-424. doi: 10.1007/s11633-006-0414-0

##### 计量
• 文章访问数:  15
• HTML全文浏览量:  18
• PDF下载量:  6
• 被引次数: 0
##### 出版历程
• 收稿日期:  2020-03-02
• 录用日期:  2020-06-30
• 网络出版日期:  2020-09-09
• 刊出日期:  2020-10-01

## DiscoStyle: Multi-level Logistic Ranking for Personalized Image Style Preference Inference

### English Abstract

Zhen-Wei He, Lei Zhang, Fang-Yi Liu. DiscoStyle: Multi-level Logistic Ranking for Personalized Image Style Preference Inference. International Journal of Automation and Computing, 2020, 17(5): 637-651. doi: 10.1007/s11633-020-1244-1
 Citation: Zhen-Wei He, Lei Zhang, Fang-Yi Liu. DiscoStyle: Multi-level Logistic Ranking for Personalized Image Style Preference Inference. International Journal of Automation and Computing, 2020, 17(5): 637-651.

/

• 分享
• 用微信扫码二维码

分享至好友和朋友圈